Hey there,

You've probably heard me mention this before: Agent building lets you bottle your expertise. You build with your expertise, process, and taste, and it's immediately differentiated from an LLM alone.

But now people are talking about skills, vibe-coding, agents... which of these should you be building with?

Here's how I think about it.

None of this is actually new. The forms expertise can take just changed shape. The consulting project became the agent. The short "how to buide" became the skill. The custom-built software became the vibe-coded app. Same underlying job — packaging what you know so other people can use it — new containers.

There are three containers worth knowing about right now. Each one fits a different shape of expertise and a different shape of user.

Skills

A skill is a small, sharp piece of judgment that someone else drops into their existing AI workflow. Anthropic launched a format for this last fall — basically a folder with a markdown file that tells Claude "when this kind of task comes up, here's how to handle it." OpenAI adopted the same standard a couple months later.

Use a skill when your expertise is a rule of thumb, a checklist, a structural preference, or a way of approaching a specific kind of task. "Always rewrite cold email openers to lead with a question.""When summarizing a research paper, surface the methodology before the conclusions." What used to be a short guide like "5 ways to write cold emails with great open rates".

There's a good opportunity to post these on Github and social media. Build and share these for free to demonstrate credibility and build a following, if that's something you're trying to do.

Agents

An agent is for when the user wants the outcome, not the steps.

Most people don't want to learn how to write a buyer persona. They want a buyer persona. They don't want to learn how to write a cold email that gets replies. They want a cold email that gets replies. The agent skips the teaching and just does the thing.

This is where I think most bottled expertise is going to live, because most people most of the time want results, not knowledge. It's also the layer where monetization is starting to actually work. Agent.ai is building toward it. Poe pays creators per message. The GPT Store stalled but the demand it surfaced was real.

It's the layer I'm personally betting on.

Vibe-coded apps

A vibe-coded app is still software, just significantly easier to build. A whole experience — built by one person in days instead of a dev team in months.

If you've ever wanted to build the SaaS version of your expertise, the build part is no longer the bottleneck.

But distribution is brutal. Building the app is now the easy part. Getting anyone to use it, pay for it, and stick with it is as hard as it ever was — and most expertise doesn't actually need a whole app to be useful. A skill or an agent gets you 90% of the way for most use cases.

Reserve the full app for when you've already validated that people want what you're building, and the agent form is starting to creak under the weight of what you're trying to do.

​Sam Mallikarjunan, GM of Agent.ai, has said his goal for the 2027 conference is to put 1,000 people on stage who've made life-changing money on the platform. I think they'll get there. The reason is that everyone has some kind of expertise — and now, finally, everyone has a way to bottle it. The designer will use an agent built by a storyteller. The storyteller will use one built by a data analyst. Expertise becomes transferable in a way it never was.

Which layer would you bottle your expertise at — a skill, an agent, or a whole app? Hit reply and tell me. I'm collecting these for a future issue. And, here are some of my agents below!

​Buyer Persona Builder (free)​

This is where it starts. Feed it LinkedIn profiles, and it tells you how that persona thinks, how they make decisions, what language resonates with them, what turns them off. Over 17,000 people have used it. Think of it as the foundation — you can’t write well to someone you don’t understand.

If you've used it before, this is a new and improved version.

​Cold Email Analyzer (paid)​

Take the persona output and drop in your cold email draft. It reads your email through your buyer’s eyes — shows you where you lose them, what lands, what feels generic. Then it rewrites the weak parts in their
voice. With reasoning: why this line works for this person and why that one doesn’t.

​Website Copy Analyzer (paid)​

Same idea, applied to your landing page. Section by section — headline, hero, CTAs, proof points — it tells you how your target persona would actually react. Where they’d bounce. Where they’d lean in. With reasoning and context, not just a score.

First few runs are free. Please share feedback.

Enjoy!

Shaalin